Machine learning-based assessment of grassland degradation levels on the Qinghai-Tibet Plateau from 2000 to 2020

• A novel remote sensing framework assessed grassland degradation. • Combining eco-geographical zones and space-for-time substitution method. • Degradation intensified from east to west with 13.4 % recovered and 8.7 % remained degraded. • Human activity and climate interactions dominated degradation drivers. Remote sensing (RS) technology is an indispensable tool for grassland degradation monitoring and assessment at the regional scale. However, most of these remote sensing-based grassland degradation applications rely on a limited set of common indicators (e.g., fractional vegetation coverage (FVC), net primary productivity (NPP) or aboveground biomass) that are monitored through time series, which makes these methods not comprehensive and explicit enough for direct spatial comparison. In this study, we developed a remote sensing monitoring framework for grassland degradation that integrates eco-geographical zones with a space-for-time substitution approach on the Qinghai-Tibet Plateau (QTP). Five grassland eco-geographical zones were established based on long term abiotic (climatic and geographic) and biotic (NPP) data to ensure consistent degradation assessment criteria under similar environmental conditions and comparable grassland states. By integrating field data of grassland degradation level and regional indicators (FVC, NPP as well as soil conservation (SC) and water retention (WR)), also considering the spatial heterogeneity of grasslands across different eco-geographical zones, we assessed grassland degradation on the QTP with overall accuracy reaching 0.87. We detected high spatial heterogeneity in grassland degradation levels across the QTP, with a trend of increasing severity from east to west. The average area of no, slight, moderate and heavy degradation accounted for 33.4 %, 25.3 %, 14.8 %, and 26.6 % of the total grassland area on the QTP in the 2000–2020 period. Temporal dynamics analysis revealed that 13.4 % of grassland showed recovery trends, while 8.7 % maintained stable degradation status, with the remainder demonstrating either no significant change or dynamic equilibrium. Furthermore, the key drivers influencing the spatial distribution of grassland degradation were investigated with a geographical detector model. The human activity intensity index (HAI) exhibited the highest explanatory power among individual factors (q = 0.75). Notably, the interaction between mean annual precipitation and degree-days below 18 °C (q = 0.98) emerged as the dominant factor, emphasizing the strong interaction influence of climatic variables in shaping grassland degradation patterns across the QTP. Our study introduces a method for assessing grassland degradation level considering spatial variation of grassland, which provides more explicit information for management and sustainable development of grassland ecosystems compared with traditional methods.